Estimation of Hounsfield unit conversion parameters for pelvic CT images

Area of Science:

• Medical Physics
• Radiotherapy Technology
• Image Processing

Background:

• Computed Tomography (CT) pixel values are linearly scaled to Hounsfield Units (HU) for tissue characterization.
• Inconsistencies in HU scaling parameters have been observed during CT image export from radiotherapy treatment planning systems (RTPS), especially with Radiation Therapy Oncology Group (RTOG) format.
• Accurate HU values are critical for radiotherapy treatment planning and image analysis.

Purpose of the Study:

• To evaluate methods for estimating air and water pixel values for accurate linear scaling of CT images.
• To compare different HU estimation techniques using pelvic CT data from a multi-center trial.
• To validate selected estimation methods for interpatient image registration applications.

Main Methods:

• Compared various methods for estimating water (e.g., mean, min, max, centroid values in the bladder) and air pixel values within CT images.
• Utilized pelvic CT images from a large multi-center trial and a more recent, homogeneous dataset for validation.
• Assessed accuracy against original Digital Imaging and Communications in Medicine (DICOM) export parameters.

Main Results:

• Estimating water using the mean pixel value within a bladder region of interest was more accurate than other tested methods.
• The accuracy of air estimation methods varied depending on CT scanner and RTPS characteristics.
• Selected air and water estimation methods demonstrated significantly lower error on a homogeneous validation dataset.

Conclusions:

• A method using the mean bladder pixel value for water estimation offers improved accuracy for CT image scaling.
• Robust and universally superior methods for air estimation across different systems remain challenging.
• The validated methods are suitable for linear scaling of pelvic CT images in interpatient image registration.